Personality Predictions Based on User Behavior on the Facebook Social Media Platform

TLDR

Social‑network research has produced many methods to infer users’ personalities from their online activities and language, differing in algorithms, data sources, and feature sets. This study investigates how accurately Facebook users’ Big‑Five personality traits can be predicted from various feature sets. Using the my Personality dataset, the authors compare four machine‑learning models—including XGBoost—to predict Big‑Five traits from social‑network and linguistic features and examine feature–trait correlations. XGBoost achieved the highest overall accuracy of 74.2%, and using social‑network features yielded the best extraversion prediction accuracy of 78.6%.

Abstract

With the development of social networks, a large variety of approaches have been developed to define users' personalities based on their social activities and language use habits. Particular approaches differ with regard to different machine learning algorithms, data sources, and feature sets. The goal of this paper is to investigate the predictability of the personality traits of Facebook users based on different features and measures of the Big 5 model. We examine the presence of structures of social networks and linguistic features relative to personality interactions using the my Personality project data set. We analyze and compare four machine learning models and perform the correlation between each of the feature sets and personality traits. The results for the prediction accuracy show that even if tested under the same data set, the personality prediction system built on the XGBoost classifier outperforms the average baseline for all the feature sets, with a highest prediction accuracy of 74.2%. The best prediction performance was reached for the extra version trait by using the individual social network analysis features set, which achieved a higher personality prediction accuracy of 78.6%.